Safwan Wshah
Total Page:16
File Type:pdf, Size:1020Kb
Safwan Wshah Assistant Professor 351B Votey Department of Computer Science Office: +1-(802) 656-8086 University of Vermont E-mail: [email protected] Research Machine Learning, Image and Video Processing, Deep Learning, Pattern Recognition, Interests Computer Vision, Document Imaging, Digital Signal Processing, Control Systems and Computer Architecture. Education State University of NewYork at Buffalo, Buffalo, NY USA PhD, Department of Computer Science and Engineering, May 2008 to June 2012. University of Cincinnati, Cincinnati, OH USA PhD Student, Department of Electrical and Computer Engineering, Sep 2007 to May 2008.(Transfered to State University of NewYork at Buffalo ) The University of Jordan, Amman, Jordan MSc. in Communication engineering, Department of Electrical Engineering, Sep 2002 to Jul 2005. Princess Sumaya University, Amman, Jordan BSc. in Electronics engineering, Department of Electronics Engineering, Oct. 1996 to Feb. 2001. Doctoral Title: Word Spotting in Multilingual Handwritten Documents. Dissertation Advisor: Prof. Venu Govindaraju, SUNY Distinguished Professor. A new approach has been implemented for keywords spotting in multilingual handwritten documents based on statistical Markov models. The approach is script independent scal- able over many languages such as English, Arabic and Devanagari and has many applica- tions in information retrieval and indexing including language identication of handwritten documents. Research Experience University of Vermont, Burlington, VT, USA August 2017 to Present Assistant Professor, Department of Computer Science PARC - A Xerox Company, Webster, NY, USA June 2014 to June 2017 Research Scientist Creating new machine learning and image processing algorithms in computer vision and document imaging fields for transportation, healthcare and education. • Vehicle Passenger Detection System: I Implemented scalable Deep Neural Net- works algorithms for counting passengers in vehicles. Xerox Vehicle Passenger De- tection System identifies the number of occupants in a vehicle with more than 95% accuracy, at speeds ranging from stop and go to 100 mph. I created scalable deep learn- ing convolutional neural network algorithms to count the number of passengers inside the car, for more information refer to Xerox-Vehicle-Passenger-Detection-System. • Digital Alternatives: I Implemented high adavnce image processing algorithms to analyis documents captured from different sources (mobile camera, electronic, 1 of 8 scanned, etc) in order to fill them electronically on fly. The Xerox Digital Alter- natives Tool maintain productivity and reduce document workflow complexity in an always-connected world. It is a workflow solution supporting todays increasingly mo- bile knowledge worker population, providing the ability to complete multiple workflows within a single application and without the need for paper.For more information refer to Xerox Digital Alternatives. • Xerox Ignite Educator Support System: I Implemented image processing and deep learning approaches based on auto-encoders and convolutional neural networks to recognize students handwriting from elementary schools. Ignite is a workflow and software solution that is using the power of data to transform K-12 education. Teacher would first scan students homework and/or exams into the Ignite system via a range of multifunctional input devices. Xerox Ignite reads, interprets, and analyzes the students work in minutes. For more information about the project refer to Ignite Educator Support System. • Human Video analytics: In this on-going project we are building a set of algo- rithms for human activities recognition in surveillance cameras, algorithms include detection, tracking and action recognition. • Surgical Video Analytics: This Project is a collaboration with University of Rochester Medical Center (URMC) in which we Implemented Deep Neural Networks and BoW algorithms for action quality assesment. Videos captured in-vivo during a surgical procedure are often post-analyzed in order to evaluate the quality of the procedure, identify errors that have taken place, assess the expertise and skill level of the surgeon, and to provide coaching and feedback to students of surgery. This analysis is currently done manually, requiring many hours of laborious inspection by expert surgeons. Xerox Research Center, Webster, NY, USA August 2012 to May 2014 Research Scientist Creating new machine learning and image processing algorithms in computer vision and document imaging fields for transportation, healthcare and education. • Form Registration:I Implemented image processing algorithms for independent global and local form registration for health care transaction processing. • Crowd Sourcing for Medical Forms: I Developed image processing modules to process different medical forms with emphasis on their crowdsourcability. • Statistical Toolkit: Develop high level statistical toolkit (C/C++) that implements statistical analysis and machine learning functionalities such as Neural network, con- volutional neural networks. • Expression Spotting System: I Implemented deep learning algorithms to perform a low level of information retrieval that detect and recognize specific information such as mail address, email address, phone number, dates, numerical tables, page number, etc... • Handwriting and Machine Printed Text Separation : I participated in imple- menting deep learning algorithms that separate handwritten from machine printed text in structured documents using auto-encoders. Center for Unified Biometrics and Sensors, University at Buffalo, Buffalo, NY USA Research Assistant May 2008 to Sep. 2010 • Keyword spotting in off-line Handwritten Documents : I Proposed filler and background models for keyword spotting that combines local scores and global word hypotheses scores to learn a classifier for keywords and non-keywords based on statistical Markov models.. • Arabic Handwritten Recognition: I Developed features and statistical ap- 2 of 8 proachs to recognize full Arabic handwritten documents without line segmenta- tion. The method used n-gram language modeling for enhancing the results. In this project I developed advanced technique based on continuous probability-Connected Hidden Markov Models with robust features such as GSC (Gradient-Structural- Concavity ) features to gain the best performance on full Arabic document. The developed algorithm alleviated the need for segmenting Arabic text prior to recog- nition. The algorithm performed concurrently both segmentation and recognition, the algorithm achieved 70% accuracy on the public AMA dataset. Teaching Experience Primary instructor Adjunct Professor, University of Rochester, Electrical and Computer Engineer- ing Department (Fall-2016): Teaching Digital Signal Processing course as primary instructor. The course had 45 students from graduate and undergraduate levels from Electrical and Biomedical Engineering Departments. Internal course, PARC (Fall-2015): Teaching Deep Learning in Computer Vision course for more than 50 engineers and researchers from different backgrounds. Teaching Assistant Introduction to C++, Fall-2007 and Numerical Analysis using Matlab, Spring-2008. My responsibilities included holding office hours, grading exams and homework, writ- ing and grading quizzes and assigning grades. Software Engineering, Fall-2010 and Fall-2011: My responsibilities included partic- ipating in course design, suggesting and grading projects and writing and grading exams. Computer Architecture, Spring-2011 and Spring-2012: I Handled the project part which worth one-third of the course total grades. My responsibilities included prepar- ing the project material, project lectures, assignments, supervising and helping stu- dents during office hours and grading. Mentoring During my professional career as researcher at PARC and Xerox research labs I mentored many senior graduate students during their summer internships. Select the research area, hire students with the right skills through series of interviews, setup the needed resources, guide interns on how to conduct research and help in publishing their work. Industry Experience Xerox, Webster, NY, USA Software Engineer, Internship May 2011 to Sep. 2011 Form Registration: Built a full frame work to register forms globally and locally by devel- oping machine learning pattern matching algorithm, the algorithm showed high accuracy for many popular forms such as medical HCFA and UB forms. Applied Media Analysis, Collge Park, MD, USA Software Engineer, Internship May 2010 to Sep. 2010 Research and development of software for Arabic Optical Character Recognition using continuous probability-Connected Hidden Markov Models. The developed algorithm alle- viated the need for segmenting Arabic text prior to recognition. The algorithm performed concurrently both segmentation and recognition. 3 of 8 Copanion, Andover, MA, USA Software Engineer, Internship May 2009 to Sep. 2009 Development of advanced convolutional neural network algorithms to recognize English handwritten text used for tax form recognition. Lead Technologies , Amman, Jordan Software Engineer Jan. 2001 to Sep. 2007 Development and Research at different levels: • Research in Image, Document, and Video processing algorithms. • Implementing the developed algorithms under C, C++, Java, and .Net. • Involved in training and guiding new members and updating work guidelines. • Team leader (9 Team members), managing and supervising all team projects. Master Thesis Title: A Robust Algorithm for Face Detection in Color Images